Interpretable models to explain Heterogeneous Catalysis

Mathematical descriptions of experiments hold the key to optimize industrial processes. Particularly, modeling catalytic processes from the atomistic

level to the real-life size reactors allows obtaining deep physical insights on the studied systems, which can be used to improve the activity and the selectivity of the underlying chemical process. During the last 20 years, the standard theoretical procedure was Multi-Scale modeling. This methodology has proven to be a reliable and robust technique for simple systems. However, as the complexity of the system increases, classical Multi-Scale modeling is not able anymore to reproduce the experimental trends. New computational techniques based on Statistical Learning (SL) are used to overcome this issue. Nonetheless, most part of SL procedures are not physically interpretable. In this thesis, we present a procedure to obtain accurate and physically interpretable models to predict and explain heterogeneous catalytic

systems by using a Bayesian-based algorithm. The present work is divided into three main blocks consisting of two chapters each. In the first block, the theoretical background for classical Multi-Scale modeling and for the new data-driven computational tools are discussed. Next, in the second block, applications for classical and automated Multi-Scale modeling are presented. Finally, the data-driven method developed by our group to obtain interpretable equations is applied to two different systems.

If you would like to follow the ceremony on Zoom, please register here.